The deployment of Intelligent Transport Systems in Europe [1] deals with the integration of the information and communications technologies of vehicles and road facilities. Particular attention is paid to next generation vehicles, information and communication strategies, services and facilities [2].
In this paper, we look at the likely trajectory of emerging Vehicle Ad Hoc Networks (VANETs) applications and their subjacent technology. Five technical aspects (Transmission, Routing, Quality of Service, Security and Location) that we consider are differential aspects of VANETs from current ad hoc networks were reviewed.
There are several research works related to VANETs applications, as we can see in [3,4]. Au-thors in [5] developed an application in an urban area that quantifies the lost time and pollution generated in thousands of dollars a day when a person is look-ing for a parking space. In [6] there is the example of Cooperative Collision Warning Systems, where vehi-cles were equipped with sensors, communication
hardware, wireless protocols, estimators, and colli-sion warning algorithms.
VANET applications examples reported by the lit-erature conclude that designing a VANET applica-tion is a complex process that requires analyzing all the technical aspects involved in the context where the application will be used. It is necessary to define communications systems that feature a convenient, stable, secure and economic distribution of data over highly mobile nodes. Vehicles require reliable loca-tion mechanisms and also a communications capabil-ity able to distinguish between critical and informa-tive messages.
We look into the routing aspect in depth and we consider the density of vehicles as one of the main context characteristics that affect routing protocols in VANETs.
In a concrete example we show a procedure to compute vehicle density in the streets of Xalapa city, México. The setup considers the use of acoustic sen-sors which allow the detection of the presence of a vehicle moving freely, or vehicles that are waiting in front of red lights or in a traffic jam. The system is
Journal of Ambient Intelligence and Smart Environments 6 (2014) 469–481DOI 10.3233/AIS-140271IOS Press
This article is published online with Open Access and distributed under the terms of the Creative Commons Attribution Non-Commercial License.
able to detect a cluster of vehicles moving in the street.
The main contribution of this paper is our ap-proach to determining the density level according to the signals collected from the acoustic sensors, the distance of the street and the time intervals. Once the density level has been determined, we propose that VANET applications change their routing protocols according to these density levels in a dynamic way.
The paper is structured as follows: Section 2 briefly reviews related work; Section 3 analyzes the main technical aspects that distinguish the VANETs from other ad hoc networks. We detail how the tech-nical aspects of VANETs should be involved in VANET application designs. Section 4 details how our simulation model detects the density of vehicles within urban environments based on acoustic signals. We determine a convenient routing protocol accord-ing to the vehicular density levels. The Conclusions are in Section 5.
2. Vehicular ad hoc networks
Today vehicles include in their facilities a broad variety of sensors providing the driver with informa-tion related to the vehicle’s functioning conditions, journey development and security guidance. Along with the increasing capabilities of radio communica-tions systems, it is now possible to monitor remotely vehicles’ functions and the conditions of the sur-rounding environments, such as the traffic on the road, available parking places, etc., a main focus of open interest being the best way to deliver vehicle sensor data about the vehicles or infrastruc-ture.
VANETs are ad hoc networks that have predict-able node movements due to the existing infrastruc-ture (roads, motorways, etc.) and traffic laws. How-ever, context awareness becomes difficult because the immediate context changes very fast. In fact, the high velocity and density of vehicles are clear char-acteristics that make VANETs very distinct from other ad hoc networks. The same occurs with routing and security mechanisms that have to deal with highly unstable network topologies.
VANETs have their own unique characteristics, such as:
In the physical layer, the 802.11p wireless stan-dard is the most generalized protocol for VANETs [7]. This standard, also called wireless access in ve-hicular environments, uses the multi-hop technique
for routing within a range of between 300 and1000 meters.
Network routing protocols for VANETS have been analyzed in several pieces of research: The work in [8] proposes a proactive routing where the movement of vehicles is expected and studied to im-prove performance. In [9] the authors consider the current location of vehicles to reduce the number of signal collisions in broadcast messages. In [10] the use of directional MAC protocol to reduce the colli-sion rate is proposed. This protocol takes advantage of the vehicles’ predictable movements because of the road and driving rules.
At the application level there are different ways for classifying VANET applications. For example, in [11] the authors mention four application categories: Active safety, public service, improved driving and business/entertainment. In contrast, [12] classifies applications based on their requirements of security and types of information: Vehicle-to-Vehicle Coop-erative Awareness, Vehicle-to-Vehicle Unicast Ex-change, Vehicle-to-Vehicle Decentralized Environ-mental Notification, Infrastructure to Vehicle (One-Way), Local Road Side Unit (RSU) Connection, and Internet Protocol RSU Connection. In [13], applica-tions have three main categories: Safety, traffic monitoring/optimization, and infotainment. In [14] the classification is: Life-Critical Safety Applications, Safety Warning, Electronic Toll Collection, Internet Access, RSU Service Finder, and Group Communi-cation. In this research we use this last classification.
3. Technical aspects involved in the design of
VANET applications
VANETs are a very distinct category of ad hoc networks, so, the diverging requirements of VANET applications making the design of a comprehensive
Fig. 1. VANET representation [1].
A. Reyes et al. / Vehicle density in VANET applications470
AUTHOR COPY
communication system a very complex topic. We based it on the VANET classification of Yi and Moayeri [14], which divides the applications into six groups: Group A: Life-Critical Safety Applications, Group B: Safety Warning Applications, Group C: Electronic Toll Collections, Group D: Internet Ac-cess, Group E: RSU Services Finder, Group F Com-munications.
A lot of previous work exists on communication in VANETs; however, that research is very diverse and has not yet been well categorized [11,14,15].
We therefore start with an analysis of the VANETs application requirements. We have seen that, depend-ing on the type of road, the density of vehicles, the speed of the vehicles, and so on, the communications requirements would not be the same. Traffic commu-nications are more chaotic in cities than on highways, and node density is not the same on highways as in cities or on rural roads.
So, we have categorized the VANET characteris-tics by considering a set of five main technical as-pects that differs from other ad hoc networks: Transmission, Routing, Quality of Service, Security and Location
Finally, we suggest the best configuration option for each VANET application group.
Technical aspect 1 – transmission means The com-munication standard for vehicle-to-vehicle communi-cation and vehicle-to-infrastructure communication is the Dedicated Short Range Communication (DSRC) [14] that supports Wireless Access in the Vehicular Environment (WAVE) [16]. WAVE allows data ex-change between vehicular devices in rapidly chang-ing communications environments, where mobile nodes may move up to 200 km/h and the distances between them are between 5 and 500 meters.
The reserved bands of the radio electric spectrum were divided into seven 10MHz-wide channels. Channel Ch178 was reserved for protocol control. Channels Ch172 and Ch184 were reserved for acci-dent avoidance applications and public safety com-munications. The rest, (Channels Ch174, Ch176, Ch180 and Ch182) were used for both safety and non-safety applications. Thus, the correct selection of channel for Group A applications is Ch172, but for the rest of the applications (Groups B, C, D and E) it can be any out of Ch174, Ch176, Ch180 and Ch182.
Technical aspect 2 – routing capabilities Conven-tional topology-based routing schemes are not suit-able for VANETs. Reactive routing schemes fail to discover a complete path due to frequent network
partitioning, and proactive routing protocols are overwhelmed by rapid topology changes and even fail to converge during the routing information ex-change stage [17].
Routing must also consider mobility of nodes from zero kilometres per hour (stationary vehicles), to over 140 km/h on highways. So, routing cannot be based on group addresses, as in ad-hoc networks. Routing must rely on the locations of the nodes to determine the delivery of packets.
In [18] the authors present a survey of geocast, broadcast and multicast technologies, which can be applicable to V2V networks.
Intelligent opportunistic forwarding using speed information is explored in [19]. Our proposal is to use geocast in application Groups A and B to limit the destination region of the message related to safety; to broadcast in Group F with continuous up-dates of information among neighbouring nodes; and unicast in the rest of the groups. Also, Groups A and B are specific in not involving bidirectional pro-tocols, as opposed to to Groups C and D. Group C is the only type of application where single-hop proto-cols are more adequate than multi-hop. Finally, for Groups E and F, location-based routing approaches are suggested as being the most suitable.
Technical aspect 3 – quality of service In road traf-fic scenarios, when an emergency event occurs, the number of vehicles that approach the accident may increase dramatically. This might create a spike in emergency messages in the VANET if they are broadcast by every vehicle, leading to a collapse of the channel [10]. Our proposal provides time restric-tions to ensure quality of service for the different types of application groups. On one side, we have Groups A and B, which do not have any restrictions on the response time since they use one-way no-response messages. The proposed maximum trans-mission delay is 250 ms, with a maximum jitter of 100, and a 52 Mbps bandwidth. Less restrictive are applications on Groups C, D and E. We propose for them a 2–5 second response time, 400 ms for the transmission delay, and 24–32 Kbps for the band-width. For this application group the jitter is not con-sidered critical. Finally, for Group F there is a maxi-mum response time of 1s, but with delays of up to 200 ms. a bandwidth needed of 1 Kbps is enough since the messages are short and plain-characters.
Technical aspect 4 – security levels In VANETs malicious behaviour could be fatal to other users. VANET security requirements such as: Message
A. Reyes et al. / Vehicle density in VANET applications 471
AUTHOR COPY
Non-Repudiation, Access Control, Message Confi-dentiality, Privacy and Anonymity, and Liability Identification are detailed in [14].
Group D applications require only access control and some degree of confidentiality on up-stream messages, which can be achieved with symmetric keys. We suggest extending the two previous re-quirements of Group D only with message authenti-cation and integrity for Group E. In fact, message authentication and integrity is also suggested for the rest of the groups, except for Group F, where only authentication could be requested. Special attention is given to Groups A and B, because they should be given a grant of access without confidentiality re-quirements and with no access control. Group C has a high requirement for Message Confidentiality and Liability Identification.
Technical aspect 5 – location accuracy VANET applications require different levels of accuracy re-garding location. Some applications can work with errors from 10 to 30 meters, while others, for exam-ple security distance warnings, require errors to be lower than 1 meter.
The most common solution is the Global Position-ing System (GPS); however, it is not always avail-able. A variant of GPS is Differential GPS, which calculates position with a known physical one. Map-matching techniques combine road databases with the actual position of a vehicle. It is able to estimate a trajectory by obtaining several position changes peri-odically. Dead reckoning calculates current position by knowing the initial position and data such as speed, acceleration, time, distance or direction. Ac-cording to [15], in 30 seconds, at a speed of 100 km/h, the accumulated positioning error will be from 20 to 30 meters. In [8], a data-fusion technique calculates the lane in which a vehicle is driving using neigh-bouring messages.
We propose a differential GPS solution for Groups A and B, and data fusion techniques for Groups D and F. The applications of Group C can rely on the location information provided by the local infrastructure. Finally, we propose to combine GPS and map-matching techniques for Group E.
4. Network node density impact
When facing the design of VANET applications, it is necessary to consider all the technical aspects (see Section 3) involved in the context where the applica-tion will be used.
Moreover, before the implementation and deploy-ment of any technology it is important to know where the problematic intersections are, what times of day are more or less busy, what the traffic major flows are, etc.
We will focus on technical aspect 2 (routing), which means analyzing the kind of traffic, vehicle density, the type of road, speed of vehicles, etc. Spe-cifically, we look in depth into how node density affects routing.
The VANET application design should consider on one hand that in urban roadway conditions, the number of vehicles is usually high, the inter-vehicle spacing is small, and a fixed communication infra-structure is available. However, in this case, a very high traffic density can saturate the network very easily. Authors in [20] analyze VANET routing in high node density conditions.
On the other hand, if the node density is very low, then inter-vehicle spacing can be large. There is very little or no fixed communication infrastructure avail-able. A combination of low node density and high node mobility message forwarding could make VANET communication impossible. In [17] there is a range of VANET routing protocols for low density rural areas and for describing their associated limita-tions.
In [21] authors find that protocols based on topol-ogy, such as Advanced Adaptive Gossiping, and Multi-Point Relaying tend to have a decreasing re-transmission efficiency as the node density increases, indicating they scale poorly with node density. Sto-chastic broadcast and the distance methods have thresholds that are set directly by node density.
Our simulation model determines the routing pro-tocol that may affect the VANET application design.
The model is based on the vehicular density of an urban environment at a given moment.
In the validation of the proposed models, we have employed a connection between a vehicle traffic simulator and a wireless network simulator.
The vehicle traffic simulator allows us to simulate under controlled conditions the traffic with different programmable parameters, such as the displacement velocity of the vehicles, the time that traffic lights are active, traces of the vehicles’ position within the sce-nario, etc. In addition, we can change traffic flow density at some points, and in consequence we can observe how the propagation of these conditions af-fects the traffic in other places in our scenario. In order to capture near real-world traffic behaviour, the simulated car rate follows traffic measurements that we have gathered at five conflicting points in the
A. Reyes et al. / Vehicle density in VANET applications472
AUTHOR COPY
streets and avenues of Xalapa de Enríquez city, see Fig. 2, by use of an acoustic sensor, (see in the next section the details of this sensor implementation), and finally by putting car traffic estimations into the wireless network simulator via a socket connection. In particular, the streets under consideration and their main features are:
Juan de la Luz Enríquez: A one-way avenue in the centre of Xalapa city that traverses it from east to west.
Manuel Ávila Camacho: Also a one-way street running from east to west, with a very busy intersec-tion that affects the rest of the city traffic.
Miguel Hidalgo: A one-way street that allows drivers to leave from the center towards the south.
Rafael Murillo: A two-way main road, its car flow is a continuation of Juan de la Luz Enríquez Street.
Venustiano Carranza: A two-way street that runs from east to west and vice-versa. It has a bridge which allows only one line of cars to cross at a time. There is a traffic light that controls the direction of flow over the bridge. This is a conflicting point that can cause heavy traffic jams when the traffic is heavy, which can extend along adjacent streets.
Over one week, three times a day, we systemati-cally collected traffic data (see the procedure in List-ing 1). The sensor data consists of sets of continuous traffic readings of five minutes. In the case of the conflicting intersections, we recorded the traffic read-ings on the adjacent streets by counting the car flow contributions at each branch of the intersection. In the case of the two-ways street, we deployed one sensor for each lane. In all the cases, the wireless
nodes were synchronized so that all the sensors started and finished the collection of traffic data at exactly the same time.
The wireless network simulator allows us to simu-late the adapting wireless routing protocol behaviour according to the conditions of the scenario under study; for instance the protocol can be switched into a specific mode when the observed car traffic density changes to a determined specific distribution, or when the wireless communication Quality of Service (QoS) scales up or down to a certain pre-pro-grammed value, at packet level or received signal strength indicator (RSSI) figures for example. In fact, this last mentioned feature can only be handled by a simulator that implements any sort of loss wireless communication propagation model.
The next section shows the technology used to de-tect the number of stationary vehicles with their en-gine on, the number of vehicles driving individually, with no traffic surrounding them, and the number of times that vehicles drives in a traffic flow.
4.1. Acoustic sensor signal processing
In this section we provide technical details about the implementation of our system to detect the den-sity of vehicles within urban environments. The sys-tem utilizes car engine sounds to determine the oc-currence of three movement events: A single car passing, a stationary car, and a flow of cars moving along the street. Roughly, the system is made up of four modules: Filtering, signal energy computing, event detection, and car counting [22–24].
Our experimentation procedure followed an in-cremental approach; therefore, in order to achieve predictability of the system module’s behaviour; we carried out the experiment’s trials on a portable com-puter with a microphone attached; we collected car engine sounds of the car movement events previously mentioned and recorded them onto the computer hard drive. In order to process the signal sounds with sig-nal processing algorithms at different resolution de-grees, we collected the sound signals at the highest performance rates permitted by the computer hard-ware. The sound recordings were ten seconds, the samples were 16-bits, and the sounds were acquired at a data rate of 44100 samples per second. We started the experiment by feeding the filtering stage with the recorded sound signals; output was passed to the signal energy module. Afterwards, we adjusted the parameters of the event detection module, whose output fed into the car counting module. In Figs 3, 4,
Fig. 2. City map of Xalapa, México.
1. The initial collection time was es-
tablished.
2. The program was executed to acquire
data during a period of 5 minutes.
3. The program stopped for a delay pe-
riod of 5 minutes.
4. The sequence was repeated as from
step 1.
Listing 1. Procedure for collecting traffic data.
A. Reyes et al. / Vehicle density in VANET applications 473
AUTHOR COPY
and 5 respectively are plotted recordings of the car events: one single car passing, one stationary car, and one flow of cars moving along a street.
In the filtering stage, the sound signal was proc-essed by a 40-tap Hamming band pass filter. Essen-tially, the purpose of the filter was to reject the signal frequencies that did not belong to the car engine sounds, which were acoustic signals within the range of the 50 Hz and 5 KHz band frequency; this filter implementation was realized by the use of the fol-lowing relationship:
0 1 1k k k N n N
y b s b s b s− −
= + + +�
where ky is the filter output,
0 1{ , , , }
Nb b b… is the
filter taps set and 1
{ , , , }k k k Ns s s
− −
… is a 1N + length
vector of the acoustic input samples .
ks Figure 6
shows the vehicles moving along a street signal fil-tered by our filter implementation.
To compute the vehicles’ signal energy we use the next relationship [22]:
k k ke s s= −
where ks is the current sound sample and ‚
ks Ú is the
current estimated average value of the sound signal. In the case of an embedded computer, this represen-tation is more convenient because the settling of any threshold or the searching of outliers in the signal avoids the implementation of complex zero crossing algorithms and occurs in the analysis of only positive numbers. Figure 7 shows the computation of the sig-nal energy of the acoustic signal of vehicles moving along a street.
In our implementation each one of the deployed wireless nodes sensed the sound signals while look-ing in the sound signal energy for subsets that con-tained substantial variations that we called events,
Fig. 3. Raw acoustic signal of one single car moving along a street.
Fig. 4. Raw acoustic signal of one stationary car.
Fig. 5. Raw acoustic signal of vehicles moving along a street.
Fig. 6. Filtered acoustic signal vehicles moving down the street.
A. Reyes et al. / Vehicle density in VANET applications474
AUTHOR COPYand that our algorithm related to one of the states of
the car engine signals recorded at the recollection procedure. From a distributed system approach, the computation of events locally was more efficient than the transmission to a central host, because this can significantly reduce the number of data transmissions by up to 90% [21].
In the literature there have been reported a large number of event computing algorithms [22,23]. In [24] it was shown that in many cases the scenario strongly determines the selection of the event com-puting procedure, which can lead to better outcomes. In particular it was shown that in a one-car counting system the threshold-based algorithms work more effectively than standard deviation rate algorithms [22,23]. In the threshold approach two amounts were computed, a signal energy reference and an outlier, which were intensively analyzed one against the other to get an estimation of the occurrence of vehi-cle presence; in particular we observed the behaviour of the following energy reference and outlier values: 1) energy average and the energy standard deviation, and 2) energy average and the energy variance. In Fig. 8 are shown the computations of the average and variance values of the sound signal energy of the one car moving along a street signal. In Fig. 9 are plotted the computations of the sound signal energy average and standard deviation values of the one car moving along a street signal.
From Figs 8 and 9 we observe that in some places the sound energy average stands below the both the variance and standard deviation values when there are no significant changes in signal energy. More specifically, in the first case, see Fig. 8, we can ob-serve that the signal energy average values are al-
ways kept below the energy variance even when sig-nificant changes are observed in the energy average estimations. Whereas in the case presented in Fig. 9, when the signal energy changes little, there are ob-served some crossovers between the energy average and the standard deviation estimations, which are due principally to ambient noise. On the other hand, we can see a large number of crossovers when the changes of the energy average are present, as seen in Fig. 10, where it is shown that all the energy average crossovers observed occur within ten-second time periods.
In this manner, we adopted the number of times the energy average crossed the energy standard de-viation within a given time interval as the car detec-tion criteria.
At this point, the problem was to program the de-tector time interval that achieved the highest number
Fig. 7. Signal energy from the signal vehicles flowing along a
street.
Fig. 8. The energy average vs. energy variance.
Fig. 9. The energy average vs. energy standard deviation computa-
tions.
A. Reyes et al. / Vehicle density in VANET applications 475
AUTHOR COPY
of positive car detections and the minimum number of false detections. We followed an incremental ap-proach in the following manner: An initial time in-terval was set, running the event detector and regis-tering the number of detections observed. Whenever the number of detections observed was far from the actual number of vehicles expected for the car movement event under consideration, then the time interval was adjusted, widening or reducing it, de-pending on the output observed from the event detec-tion module, and the event detector operations were repeated. For instance, in Fig. 11 we observe detec-tions issued by the sensor when the system was fed with a signal from one single car moving. In particu-lar, the crossover number was fixed to a value of 20 ≤ detection_number ≤ 200 times within a time period of one second. In this case, we counted ten one-second time periods and seven car detections
where most of the detections happened within a two to seven-second period.
In Figs 12–15, the computed detections are plotted for time windows of two, three, four, and five sec-onds respectively. In Fig. 12, there are five windows of two seconds and the algorithm issued four detec-tions. Figure 13 shows that there were three windows of three seconds and the detector reported three cars observed. In the case of Fig. 14 there were two win-dows of four seconds, and the event detector sensed two car occurrences. Finally, in Fig. 15, there is shown two five-second windows, and the detector reported the presence of one car.
From the outcomes observed for the case of a single car moving, we can see that the five-second time window enabled the best performance of the vehicle detector. However, it should be observed that the limitations of the system car detector were
Fig. 10. All detected crosses.
Fig. 11. One-second window detection.
Fig. 12. Two-second window detection.
Fig. 13. Three-second window detection.
A. Reyes et al. / Vehicle density in VANET applications476
AUTHOR COPY
constrained to the car speeds and the distance be-tween the advancing cars. In fact, if two adjacent nodes were separated by a distance of 400 meters, the maximum speed a car could still be observed by the nodes was around 299.33 km/h. On the other hand, within an interval of five seconds a car could advance around 25 meters at a velocity of 19.95 km/h, which is the system where vehicles could be observed which were separated by a minimum distance of around 12.5 meters.
The car counter module computed sums and statis-tics, such as the variance and standard deviation, of all the car detector module outputs. The car counter module could be programmed to compute the figures of the car traffic behaviour for fixed periodic time intervals, such as seconds, minutes or hours, accord-ing to the car detector module computing time win-dow’s minimum computing time period. Rather than the granularity of the traffic estimations being in the
order of seconds, better estimations could be ob-tained if the counter module ran estimations for ex-tended time periods, for instance of one hour or more. The algorithm of this procedure is illustrated in List-ing 2.
At the end of the procedure shown in Listing 2, the next variables are cleared, and the procedure starts again.
Variables: current_cars_number
current_cars_number_avg
current_cars_number_variance
current_detector_value
kt
It should be noted that within a long enough time period, the reported partial sums provided informa-tion about the car traffic density probability histo-grams. In fact, the sum outcomes can be considered as the density probability maximums multiplied by a certain normalization factor and the bin wide is the time window the car counting system ran freely, the car detection module with a five-second time window being the minimum function density bin wide.
For instance, the expected normalized distribution of the traffic in a jammed street can be like the one shown in Fig. 16, which is the density probability observed within one hour for a time window, a bin wide of five minutes.
module output and is a number that can take only one
of two values {0, 1},
current_cars_number_avg is the average
number of computed cars per time window,
current_cars_number_variance is the vari-
ance number computed per time window.
Listing 2. Car counter algorithm.
A. Reyes et al. / Vehicle density in VANET applications 477
AUTHOR COPY
Finally, the car traffic estimations were transmitted wirelessly travelling in a multi-hop fashion along the nodes in the WSN and delivered to a data collector.
The next section explains the simulation scenario of the five selected streets from Xalapa city in México, which we mentioned in at the beginning of Section 4.
4.2. Traffic simulation and density extrapolation
There are two main types of traffic simulation: Microscopic and macroscopic, which are related to the depth of analysis to be performed. The micro-scopic simulation is the movement of each vehicle individually, requiring a large amount of data and computational resources, such as memory, computing power, storage units of data, and so on. This type of simulation provides results that give a detailed pic-ture of network performance and sensitivity of the system to specific changes in traffic.
Whereas a macroscopic simulation is based on traffic analysis from a global perspective, where traf-fic is considered continuous and its application is essentially in urban planning.
In order to generate the simulation scenario (street map) and the vehicular mobility patterns we used the Mobility model generator for Vehicular networks (MOVE), which uses an open-source micro-traffic simulator called SUMO (Simulation of Urban MO-bility) [25] for simulating our traffic scenario. In this way, we have been able to simulate realistic vehicu-lar movements such as traffic jams and stops at inter-sections. We configured the scenario based on the five streets selected in Xalapa city, México, and we considered different traffic densities to enrich the SUMO simulation. SUMO generates an output file with vehicular mobility traces. This file was put into the network simulator NS-2 [26] to obtain the final simulation results. We configured the maximum speed of the vehicles as limited to 50 km/h. The transmission range was fixed at 150 meters.
After a simulation from 9:30 to 11:30 (120 min-utes, with 24 time intervals of 5 seconds) on Venustiano Carranza Street, we counted 29 vehicles, which meant 0.24 vehicles per second.
Density factor is one of the elements that most af-fects the good performance of VANET routing pro-tocols. So, this research is proposing the VANET application design with a dynamic change of routing protocols according to traffic density changes. The next section explains how we propose to carry out the dynamic election of the protocols.
4.3. Dynamic election of the routing protocols
Several environmental factors have an effect on the communication performance of routing protocols. In urban areas, one of the most important factors is the density of vehicles. It generally depends on the hour, and may change accidentally. If vehicle density is too high (traffic jams) then nodes are too crowded in the network, hops between them would cause re-dundancy, increase the network overhead, and also create the possibility of causing load imbalance. On the other hand, low traffic density could make diffi-cult to establish a network connection.
Thus, aspect 2 (Section 3.2) for VANET applica-tion design has to take dynamic changes in road traf-fic conditions into consideration, including vehicle density.
This section analyzes how the vehicle node density affects the selection of the routing protocol. The pro-posal of this research is not only to compensate for different levels of vehicle density in reconfiguring the routing protocols, we also propose to switch dy-namically between different routing protocols when the vehicle density changes.
The five selected streets in Xalapa have highly variable vehicle densities, which depend on the time of day or night. We defined three main time inter-vals:
Time Interval1: From 00:00 hrs to 06:00 hrs Time Interval2: From 06:00 hrs to 07:30 hrs, from 09:30 hrs to 13:00 hrs and from 18:30 hrs to 00:00 hrs Time Interval3: From 07:30 hrs to 9:30 hrs and from 13:00 hrs to 18:30 hrs
We discarded the use of protocols such as the Urban Multi-Hop Broadcast protocol that does not work well in urban environments with very low den-sity. This kind of protocols requires each node to rebroadcast messages to all of its neighbours, and if there are no vehicles within transmission range, then
Figure 16. Weighted vehicle traffic density probability of an ave-
nue with heavy traffic.
A. Reyes et al. / Vehicle density in VANET applications478
AUTHOR COPY
neighbours cannot be found to maintain the seamless connectivity of the network. Such protocols require flooding the work with a very high density because the bandwidth requested for one broadcast message transmission can increase exponentially.
We have also ruled out protocols that have low communication throughput and are inefficient in quickly finding and maintaining routes in VANETs environments such as: The Ad-hoc On-demand Dis-tance Vector, and the Dynamic Source Routing. Likewise, cluster-based routing protocols such as Clustering for Open IVC Networks did not work well. These kinds of protocols can achieve good scal-ability for large networks, but have some problems in fast-changing VANET systems due to the delays and overheads involved in forming and maintaining these clusters.
Since our scenario simulated streets in an urban environment, we also discarded protocols such as the Broadcomm protocol that was designed for highway networks.
In order to design VANET applications such as cooperative crossing at intersections, the possible protocols to be used in the five analyzed streets are:
High density: The Greedy Perimeter Coordination Routing works with street maps. In this protocol street crossings are the only points where packets are forwarded. This protocol is useful in urban environ-ments with high vehicles densities. The Anchor-based Street and Traffic Aware Routing uses a street map to compute the sequence of street crossings while incorporating traffic awareness.
Low density: The Greedy Perimeter Stateless Routing bases forward decisions on location informa-tion. It combines the Greedy Routing with Face Routing by using the Face Routing to avoid the local minimum where Greedy fails. This kind of protocol works best with low vehicle densities, but a mini-mum number of vehicles is necessary to be able to do Face Routing. In high densities this protocol can in-duce routing loops for Face Routing. History-enhanced Vector-based TRAcking Detection is a message broadcasting protocol based on position and movement information. This protocol has routing overhead, as long as the forwarding nodes are se-lected in every hop.
The next protocols support high and low vehicle densities: Geographic Source Routing assumes the aid of a static street map. It combines Geographic Routing and topological knowledge from street maps.
Abiding geocasts is where the packets need to be delivered to all nodes that are at some time
during their lifetime inside the geocast destination region.
Most papers focus on how to modify existing pro-tocols to better fit a VANET. We do not change the protocol configuration in order to solve the problem of continuous changes on traffic density. However we propose to design VANET applications based on a dynamic election of routing protocols according the changes on traffic density.
Our dynamic selection algorithm explains how to change of routing protocols in the five analyzed streets. The algorithm is illustrated in Listing 3.
5. Conclusions
This research focuses on which protocols and technologies would best fit VANET applications. We discussed the advantages and shortcomings of the
While (DENSITY RDi > 0,5) //high density, more than 60 vehicles per //hour at //a specific point Apply (Greedy Perimeter Coordination Routing // Anchor-based Street_Traffic Aware Routing)
While (DENSITY RDi <= 0,25) //low density, less than 30 vehicles per hour //at a/specific point Apply (Greedy Perimeter Stateless Routing, History-enhanced Vector-based TRAcking)
Where s = source, d = destination, p = packet RDi street distance, where
i = 1. Juan de la Luz Enríquez: 0.4 km i = 2. Manuel Ávila Camacho: 2.4 K i = 3. Miguel Hidalgo: 0.56 km i = 4. Rafael Murillo: 1.33 km i = 5. Venustiano Carranza: 0.7 Km
ITj time interval, where j = 1. 00:00 to 06:00 j = 2. 06:00 to 07:30, 09:30 to 13:00, 18:30 to 00:00 j = 3. 07:30 to 09:30, 13:00 to 18:30
Listing 3. Dynamic election of routing protocols according to
vehicle density changes.
A. Reyes et al. / Vehicle density in VANET applications 479
AUTHOR COPY
different alternatives of five main technical aspects (transmission, routing, quality of service, security and location), which are very significant to VANETs, and differential from Ad-Hoc Networks. The analysis was extended to six different groups of VANET applications (Life-Critical Safety Applications, Safety Warning, Electronic Toll Collection, Internet Access, Group Communication and RSU Service Finder). The best alternatives for each pair (technical aspect and application type) were suggested. The analysis demonstrated that a third dimension to the study became necessary: The environmental con-ditions. The paper extended the research in one topic that directly affects routing: Network node density.
Network node density is the number of vehicles driving concurrently on a road. It has a great impact on the performance of VANET networks by influ-encing factors such as capacity, routing efficiency, delays, and robustness. Using a practical approach, we analyzed traffic densities via a two-way method-ology; 1) we computed densities from actual traffic data provided by an acoustic sensor network with nodes deployed at few locations in an urban scenario in the Xalapa municipality, 2) we employed the com-puted local densities to feed an urban traffic simula-tor extending the density data to neighboring streets. With the proposed method we were able to classify the traffic density on three levels with a high reliabil-ity. It was shown that the performance results of the acoustic detectors came from identifying car sounds such as: A stationary car, a single car that was mov-ing and a flow of cars moving.
With factor density, making decisions about the best routing protocol to use in a VANET application becomes an easy configuration parameter that allows the dynamic change of routing protocol when vehicle density changes.
Acknowledgments
The Angélica Reyes and Cristina Barrado work was partially supported by the Spanish Ministry of Education and Science under the projects TIN2010-20140-C03-01 and TIN 2010-18989.
The Marco López and Cora Excelente work was partially supported by the Councils of Science and Technology of Mexico and the government of Veracruz State by the project number: FOMIX-VER 2010-04-152366.
References
[1] B Directorate-General for Mobility and Transport of the
European Commission, Intelligent Transport Systems in Ac-tion, EU Bookshop, 2011.
[2] C. Popescu, Case study: project civitas-success, in:
MMES’10 Proc. of the 2010 International Conference on Mathematical Models for Engineering Science, 2010.
[3] V. Gradinescu, C. Gorgorin, R. Diaconescu, V. Cristea and
L. Iftode, Adaptive traffic lights using car-to-car communi-
cation, in: Proc. IEEE 65th Vehicular Technology Confer-ence, Dublin, Ireland, April 2007, pp. 21–25.
[4] H. Noshadi, E. Giordano, H. Hagopian, G. Pau, M. Gerla
and M. Sarrafzadeh, Remote medical monitoring through
vehicular ad hoc network, in: Proc. of IEEE WiVeC 2008,
Calgary, Canada, September 2008.
[5] Spanish association of car parks and garajes, http://www.
asesga.org/, accessed May 2013.
[6] R. Sengupta, S. Rezaei, S.E. Shladover, D. Cody, S. Dickey
and H. Krishnan, Cooperative collision warning systems:
concept definition and experimental implementation, Jour-nal of Intelligent Transportation Systems 11(3) (July 2007),
143–155.
[7] D. Jiang and L. Delgrossi, IEEE 802.11p: towards an inter-
national standard for wireless access in vehicular environ-
ments, in: Vehicular Technology Conference, 2008, VTC Spring 2008, IEEE, May 2008, pp. 2036–2040.
[8] H. Menouar, M. Lenardi and F. Filali, Improving proactive
routing in VANETs with the MOPR movement prediction
framework, in: Telecommunications 2007, 7th International Conference on ITS, June 2007, pp. 1–6.
[9] R. Fracchia, Knowing vehicles location HELPs avoiding
broadcast packets storm, percomw, in: Fourth IEEE Inter-national Conference on Pervasive Computing and Commu-nications Workshops (PERCOMW’06), 2006, pp. 118–123.
[10] H. Menouar, F. Filali and M. Lenardi, A survey and qualita-
tive analysis of mac protocols for vehicular ad hoc networks,
(July 2008), 2838–2849, Mobility protocols for ITS/
VANET.
[16] K. Sjöberg, E. Uhlemann and E.G. Ström, Medium access
control in vehicular ad hoc networks, in: IEEE VTS Work-shop on Wireless Vehicular Communications, Halmstad
University, Sweden, October 2010, http://www2.hh.se/staff/
kasj/MAC_in_VANET.pdf, accesed May 2013.
A. Reyes et al. / Vehicle density in VANET applications480
AUTHOR COPY
[17] M. Zhang and R.S. Wolff, Routing protocols for vehicular
ad hoc networks in rural areas, IEEE Communications Magazine (November 2008).
[18] W. Chen, R.K. Guha, T.J. Kwon, J. Lee and Y.-Y. Hsu,
A survey and challenges in routing and data dissemination
in vehicular ad hoc networks, Wireless Communications and Mobile Computing 11(7) (October 2009), 787–795.
[19] J. LeBrun et al., Knowledge-based opportunistic forwarding
in vehicular wireless ad hoc networks, in: IEEE VTC,
Spring 2005.
[20] M. Torrent-Moreno et al., Packet forwarding in VANETs, the complete set of results, Technical Report 2006-2, ISSN
1432-7864, Dept. of Computer Science, University of
Karlsruhe, Germany.
[21] M. Slavik and I. Mahgoub, On the scalability of wireless
multi-hop broadcast protocols with respect to density in
VANET, in: International Conference on Communications and Information Technology (ICCIT), DOI: 10.1109/
ICCITECHNOL.2011.5762702, 2011, pp. 92–95.
[22] J. Polastre, R. Szewczyk and D. Culler, Telos: enabling
ultra-low power wireless research, in: Proc. of the 4th
International Symposium on Information Processing in Sen-sor Networks IPSN, Los Angeles, California, USA, ACM,
2005, pp. 364–369.
[23] L. Gu, J. Dong, P. Vicaire, T. Yan, L. Luo, A. Tirumala,
Q. Cao, T. He, J.A. Stankovic, T. Abdelzaher and
B.H. Krogh, Lightweight detection and classification for
wireless sensor networks in realistic environments, in: Proc. of the 3rd International Conference on Embedded Net-worked Sensor Systems, SenSys’05, San Diego, California,
USA, ACM, 2005, pp. 205–217.
[24] M.A. López Trinidad, C.B. Excelente Toledo and M. Valle,
Indoors and outdoors acoustic event detection for embedded
wireless sensors, in: Embedded Systems and Wireless Tech-nology: Theory and Practical Applications, Science Pub-
